Table 2. Reference-independent alignment quality measured by TM-score and GDT on our in-house test set.

GDT is scaled to [0, 1]. DRNF is trained by DeepAlign-generated alignments and uses Viterbi to build alignments.

	CNFpred		HHpred-global		HHpred-local		DRNF
	TMscore	GDT	TMscore	GDT	TMscore	GDT	TMscore	GDT
(0,1]	0.469	0.383	0.415	0.338	0.341	0.290	0.525	0.432
(0.45,0.55]	0.320	0.244	0.232	0.176	0.156	0.131	0.380	0.294
(0.55,0.65]	0.426	0.331	0.374	0.290	0.283	0.234	0.493	0.389
(0.65,0.8]	0.554	0.465	0.512	0.427	0.442	0.377	0.610	0.515
(0.8,1]	0.709	0.635	0.691	0.614	0.673	0.596	0.723	0.648